Sergo Ordzhonikidze Russian State Geological Prospecting University, Moscow, Russia:

E. A. Obornev, Head of Chair, Candidate of Physico-Mathematical Sciences, ObornevEA@mail.ru
M. I. Shimelevich, Head of Laboratory, Candidate of Physico-Mathematical Sciences
A. A. Nikitin, Professor, Doctor of Physico-Mathematical Sciences

Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow, Russia:

I. E. Obornev, Senior Researcher, Candidate of Physico-Mathematical Sciences

New approaches to interpretation of magneto-telluric sounding data (geophysics, geolectrics) are discussed in the article. The authors put forward an approximation method to processing the measured data (inverse geophysical problem) using a mathematical operator named in the modern information theory as the neural network (NS). It is shown that the constructed and learnt NS approximates the inverse operator and, finally, produces an electron simulator of a master chart or NS master chart. The formalized 2D/3D NS master chart is independent of skills of an interpreter and can be used many times with various measured data. An NS master chart is learnt in the process of interpretation, which needs modern computational packages and concurrent programming methods. On the other hand, the speed and portability of the complete NS master chart enables its use on customary notebooks. The solving time is a few seconds irrespective of the dimensionality of 2D/3D input data. The article gives examples to show that additional learning of NS master chart allows refinement of output, reduction of error and elimination of shortages of the primary learning (similarly to knowledge accumulation). Efficiency of the specific NS master charts is demonstrated in terms of 2D/3D solutions of inverse problems using model and field data obtained by the magneto-telluric sounding. The approximation neural network method and its modifications enable formal stable approximated solution of 2D/3D inverse coefficient problems in geolectric class of block models with the practically admissible accuracy without the first approximation setting. The number of the definable parameters of a medium reaches ~n·10³. The method and the complete NS master charts can be used in the in-situ express interpretation of data with a view to evaluating operation quality and adjusting surveying procedures after getting the first results.

The study was carried out using computer power of the Interbranch Super Computer Center of the
Russian Academy of Sciences. The study was supported by the Russian Science Foundation, Project
No. 14–11–00579.

1. Dmitriev V. I. Inverse Problems in Geophysics. Moscow : MAKS Press, 2012. 340 p.
2. Raiche A. A pattern recognition approach to geophysical inversion using neural nets. Geophysical Journal International. 1991. Vol. 105, Iss. 3. pp. 629–648.
3. Nikitin A. A., Cheremisina E. N., Malinina S. S. Neuro net modeling of occurrence depth of geological boundaries in the whole set of geophysical fields. Strategy of Geological Exploration of Mineral Resources: Present and Future (MGRI–RSGPU is 100) : International Scientific–Practical Conference Proceedings. Moscow : NPP Filtrotkani, 2018. Vol. 1. pp. 408–409.
4. Agbash I. A., Sobolev A. Y. Fast analogues of direct and inverse problems of Russian laterlog by neural network technology. Interexpo GEO-Siberia. 2016. Vol. 2, No. 1. pp. 196–200.
5. Khusainov A. T. Use artificial neural networks for operation estimation of oil fluids. Scientific forum. Siberia. 2016. No. 3. pp. 69–71.
6. Moskowsky I. G., Balaban O. M., Fedorova O. S., Kochetkov A. V. Application of neural networks at interpretation of these electromagnetic sounding of vertically jointed environments. Internet-journal “Naukovedenie”. 2015. Vol. 7, No. 1(26). Available at: https://naukovedenie.ru/PDF/05TVN115.pdf (accessed: 25.10.2018).
7. Shodiq M. N., Kusuma D. H., Rifqi M. G., Barakbah A. R., Harsono T. Neural Network for Earthquake Prediction Based on Automatic Clustering in Indonesia. International Journal on Informatics Visualization. 2018. Vol. 2, No. 1. pp. 37–43.
8. Jiang Fei-Bo, Dai Qian-Wei, Dong Li. Nonlinear inversion of electrical resistivity imaging using pruning Bayesian neural networks. Applied Geophysics. 2016. Vol. 13, Iss. 2. pp. 267–278.
9. Raj A. S., Oliver D. H., Srinivas Y. An automatic inversion tool for geoelectrical resistivity data using supervised learning algorithm of adaptive neuro fuzzy inference system (ANFIS). Modeling Earth Systems and Environment. 2015. Vol. 1. DOI: 10.1007/s40808–015–0006–5
10. Cranganu C., Luchian H., Breaban M. E. Artificial Intelligent Approaches in Petroleum Geosciences. Cham : Springer, 2015. 290 p.
11. Shimelevich M. I., Obornev E. A., Obornev I. E., Rodionov E. A. Numerical methods for estimating the degree of practical stability of inverse problems in geoelectrics. Izvestiya. Physics of the Solid Earth. 2013. Vol. 49, No. 3. pp. 356–362.
12. Shimelevich M. I., Obornev E. A., Rodionov E. A., Obornev I. E. The neural network approximation method for solving multidimensional nonlinear inverse problems of geophysics. Izvestiya. Physics of the Solid Earth. 2017. Vol. 53, No. 4. pp. 588–597.
13. Tikhonov A. N., Arsenin V. Ya. Methods of Ill-Posed Problem Solution. 2^nd Edition. Moscow : Nauka, 1979. 288 p.
14. Feldman I. S., Okulessky B. A., Suleimanov A. K., Nikolaeva V. I. Kuncherov V. A., Chamo S. S. MTS exploration within the work package of regional hydrocarbon prospecting in the European Russia. Zapiski Gornogo instituta. 2008. Vol. 176. pp. 125–131.